PulseAugur
中
实时 09:31:02
English(EN) AttnCompress: Dynamic Attention-Guided Trajectory Compression for Software Engineering Agents

新的AttnCompress框架大幅降低AI代理的上下文成本

作者 PulseAugur 编辑部 · [1 个来源] ·

研究人员开发了AttnCompress，一个新颖的框架，旨在动态压缩自主软件工程（ASE）代理的交互轨迹。该方法解决了由代理生成的长日志造成的瓶颈问题，这些日志会给上下文窗口限制带来压力并增加成本。AttnCompress利用结构感知分割、代理注意力权重进行相关性估计以及动态滚动窗口，以保留关键任务证据，同时显著减少令牌消耗和相关费用。在SWE-Bench-Verified和Multi-SWE-Bench上的评估表明，与现有方法相比，AttnCompress提高了通过率并降低了成本，展示了其模型无关和语言泛化的能力。 AI

影响 降低软件工程任务中AI代理的成本并提高效率。

排序理由 该集群包含一篇详细介绍AI代理新框架的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →

新的AttnCompress框架大幅降低AI代理的上下文成本

本文如何被排名

Signal score
13 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
该集群包含一篇详细介绍AI代理新框架的研究论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

  1. arXiv cs.AI TIER_1 English(EN) · Zhengran Zeng, Yixin Li, Rui Xie, Wei Ye, Shikun Zhang ·

    AttnCompress：面向软件工程智能体的动态注意力引导轨迹压缩

    arXiv:2609.08318v1 Announce Type: cross Abstract: The transition from human-centric assistance to Autonomous Software Engineering (ASE) agents has enabled the resolution of complex real-world SE tasks. However, the trial-and-error nature of these agents generates lengthy interact…

相关实体

相关话题